Estimating sensory delays to primate M1: a comparison of
peri-stimulus time histograms and coherence phase-frequency regression
Average PSTH
Boubker Zaaimi, Demetris S. Soteropoulos, Elizabeth R. Williams & Stuart N. Baker
Institute of Neuroscience, Newcastle University, UK. E-mail: [email protected]
Introduction
277.3/KK29
Experimental Results
50
0
-20
50
100
150
200
0
Time after stimulus (ms)
Coherence
0.06
50
100
Time after stimulus (ms)
Coherence
0.5
0.04
0.02
0.4
0.3
0.2
60
80
100
0.02
0
0
10
20
30
Frequency (Hz)
40
50
0
40
60
Frequency (Hz)
Coherence Phase
80
0
100
20
40
60
80
Frequency (Hz)
100
Coherence Phase
3p
Phase (rad)
2p
p
2p
p
40
60
Frequency (Hz)
80
100
40
60
80
20
30
Frequency (Hz)
40
50
0
20
40
Time(ms)
60
80
100
Probability per bin
Results from simulations with the
same response onset latency, but
with increasing response
duration.
The latency estimated from the
coherence phase-freqeuncy
regression (arrowheads) was
always close to the middle of the
response.
0.03
0.02
0
0.01
0
Spike train simulated as
2
inhomogeneous gamma process
(order parameter = 4)
-20
0
20
40
60
80
100
Spike Response
0.01
0.018
Fraction of intervals
0.006
0.004
0.002
0.014
50
100
150
Inter-stimulus interval (ms)
200
0.03
0.02
40
Time(ms)
60
40
Time(ms)
60
80
100
0.03
0.02
0.01
0
-20
0
20
80
100
Physiological responses
often include an initial
brief facilitation, followed
by a long lasting, but
weak, suppression (e.g.
the experimentally
recorded example cell A).
In a simulation mimicking
this, the coherence delay
estimate was
substantially larger than
the onset latency, as it
was affected by the
suppression.
Using coherence phase-frequency relationships to estimate
delay measures the average delay of the whole response,
not just the onset latency. This must be remembered when
comparing coherence delays with measurements in the
literature, which are usually based on the earliest response
to a stimulus.
In some circumstances, coherence delay estimates may be
more functionally relevant than those based on onset
latency.
0.01
0.01
0
-20
0
20
40
0.006
60
80
100
References
Time(ms)
0.002
0
0.04
Probability per bin
Spike train
20
Conclusion
Time(ms)
0.008
100
0.02
0.01
Probability changes sum to give
instantaneous spiking probability
Stimulus
80
0.04
0.01
0.04
100ms
60
0.02
0.04
Fixed change in spiking probability
follows each stimulus
0
40
Time(ms)
When the response
included both a
facilitation and
suppression, coherence
delay estimate was close
to the centroid of the
absolute response profile
(i.e. Ignoring its sign).
-20
Simulation Methods
0.02
20
0.03
0
-20
Firing Probability
per 1ms bin
0
0.04
100
Frequency (Hz)
Delays estimated from coherence phase-frequency slopes are substantially
higher than from PSTH onset latency: in the examples above 18 vs 10 ms
for the median nerve stimulation, and 17 vs 10 ms for the digital nerve
stimulation.
We investigated this further using simple computer simulations.
0
-20
0
20
0
Poisson stimulus train
10
0
0.04
Probability per bin
20
0
0
0.02
0.03
-p
0
-p
-2p
-3p
0
-p
-2p
-3p
0
Phase
p
0.03
0.01
Probability per bin
20
Phase (rad)
0
Coherence was greater at low
frequencies, reflecting the low-pass
filtering consequences of the
response duration.
For a more complex
response profile, the
coherence delay
estimate was close to
the centroid (centre of
mass) of the response.
0.04
Phases differed by p radians
between facilitation and
suppression. Regression slopes
implied a delay of 14.5ms
(arrowheads on PSTH).
0.04
0.1
0
Fraction of intervals
40
Coherence
Coherence
0
1s
20
Time(ms)
Poisson stimulus train
1 mV
0
0
0
3p
Microelectrode penetrations into primary motor cortex of a macaque
monkey, sedated using ketamine/medetomidine. Single unit activity
isolated.
Electrical stimulation of median nerve at the arm (biphasic stimuli,
0.2ms per pulse, intensity at motor threshold), or digital nerve of
index finger (10mA)
1
Stimuli delivered as Poisson stimulus trains , mean rate 10Hz.
0.002
20
Phase (rad)
!
Counts
Counts
40
Methods
!
100
60
In this poster, we examine this assumption.
!
Peri-stimulus Time Histogram
Examples of delay estimated from coherence phase for more
complex response profiles. Each plot shows the PSTH, and
the coherence delay estimate (arrowhead).
Probability per bin
80
Coherence
It is commonly assumed that coherence phase analysis will yield
similar delay estimates to response latencies, measured using timedomain methods such as the peri-stimulus time histogram (PSTH).
B
Peri-stimulus Time Histogram
Coherence
For a system with a fixed delay, phase (q) is linearly related to
frequency (f) with a slope related to the delay (t):
q(f) = 2ptf
Spike probability per bin
Coherence is a measure of correlation in the frequency domain
between two signals.
Coherence phase indicates the average phase difference between the
signals at a given frequency.
Digital nerve stimulation
Median nerve stimulation
A
Example simulation results where
cell responded with a brief
facilitation (red) or suppression
(blue) of its firing (onset latency
9.5ms, duration 10ms)
Probability per bin
PSTH
0.004
Coherence analysis is widely used to investigate oscillatory
phenomena in the nervous system.
0
0
50
100
150
Inter-spike interval (ms)
200
1. Miller S, Clark J, Eyre JA, Kelly S, Lim E, McClelland VM, Mc Donough S & Metcalfe AV. (2001). Comparison of spinal
myotatic reflexes in human adults investigated with cross-correlation and signal averaging methods. Brain Res 899, 4765.
2. Baker SN & Gerstein GL. (2000). Improvements to the sensitivity of gravitational clustering for multiple neuron
recordings. Neural Comput 12, 2597-2620.